Spectral unmixing using sparse and smooth nonnegative matrix factorization

Hyperspectral unmixing is a process to extract the endmembers and corresponding abundances from hyperspectral data. In this paper, we propose a new unmixing model based on nonnegative matrix factorization. The sparseness and smoothness properties of the abundances matrix are also taken into account. Particularly, the sparseness property is formulated by a parabolic function, and the smoothness property is expressed by the total variation norm. Furthermore, in order to verify the validity of our model, we conduct some experiments on the Cuprite data, and compare our model with some outstanding methods. The results demonstrate that our method is remarkable.